← All projects

Flexible Robotic End Effector

A three-finger, six-motor gripper whose fingers are pulled by tendons, built to pick up 5 to 20 kg automotive parts without a custom tool for every shape.

CompanyHIROTEC America, R&D

RoleProject Champion, sole owner

TimelineMay – Aug 2026

ToolsSiemens NX, CadQuery, Arduino, MATLAB, Python

VideoWalkthrough of the three-finger design.

01 / PROBLEM

One tool per part doesn't scale

HIROTEC is a Tier 1 supplier of body-in-white tooling and automation. Its robots move stamped panels, brackets, and assemblies, and almost every new part shape gets its own end-of-arm tool. That costs design hours and fixture budget every time a program changes.

I proposed a gripper that adapts instead: fingers that wrap around whatever they're given, sized for parts from 5 to 20 kg. I pitched it, secured $11,200 in funding, and ran it as the sole project champion from requirements through verification, directing a fellow intern, Noah Ericson, along the way.

02 / CONCEPT

Borrowing from the human hand

Human fingers have no motors in them. Muscles in the forearm pull flexor tendons, and small ring-shaped annular pulleys keep those tendons close to the bone so the pull turns into torque at each joint. The gripper copies that layout: the motors stay in the base, and Bowden cables carry the pull up to the fingers.

At the motor end, each tendon wraps a capstan drum turned by an Arduino-controlled NEMA 17 stepper. By the capstan equation, holding force grows exponentially with the number of wraps, so a small motor can hold a heavy load. The tradeoff is that more wraps put more stress on the line.

Hand sketch of a bent finger with tendons running over pulleys at the joint down to two grooved capstan drums — **Fig. 1**Early concept: tendons over joint pulleys, down to capstan drums with V-grooves.

Handwritten free body diagram of a finger joint with a torque balance equation — **Fig. 2**First torque balance at the joint. Measuring the moment arm from the finger's centerline instead of the joint threw this version off, so the joint moved to the center.

03 / DESIGN

Three fingers, six motors, two joints each

I modeled the gripper in Siemens NX and built the capstan drums as parametric models in CadQuery (Python), so drum diameter and groove count could change without redrawing. For each finger I derived fingertip kinematics and per-joint gravity torques using homogeneous transforms and rotation matrices.

Two joints per fingerMore joints add flexibility but split the tendon force across more load paths. Two joints gave the best balance.
Antagonistic pairsA tendon on each side holds the finger stiff in both directions. With a single tendon, the finger springs back the moment the load comes off.
Capstan driveHigh holding force from a small stepper. The catch is line wear, which made the choice of material important.
Radial railsFor the third phase, each finger rides a lead-screw rail so the grip can open or close around different part sizes.

CAD model of a single two-joint robotic finger with a rounded fingertip — **Fig. 3**Two-joint finger with fingertip pad.

CAD model of a stepper motor mount with a grooved capstan drum on the motor shaft — **Fig. 4**Stepper mount with a grooved capstan drum.

CAD render of three robotic fingers mounted on a round base plate — **Fig. 5**Fingers arranged on the base plate.

CAD render of the gripper plate from above showing three lead-screw rails converging on a central bevel gear — **Fig. 6**Rail concept. A central drive moves all three fingers in and out along their rails.

04 / PROTOTYPE

Print it, pull on it, find out

The first printed finger (Rev 1) could lift a phone when we pulled on its line by hand, and it showed us two things. Fluorocarbon line stretched under load, so we switched to braided line. More importantly, the line crossed the joint almost in line with the finger, so most of the pull went straight through the finger instead of turning it.

The torque depends on how far the line sits from the joint where it crosses it. That one finding drove every revision after Rev 1.

From there we printed four finger designs to compare routing strategies head to head. Rev 2 moved the joint down and hollowed out the finger for a better pull angle. Rev 3 wrapped the line around the joint itself. Rev 4 kept the line inside the finger, running along both sides and around the tip, so it would never touch the part being gripped.

Hands holding a black 3D-printed finger that is lifting a smartphone by its edge — **Fig. 7**Rev 1 lifting a phone, actuated by hand.

Four black 3D-printed finger designs side by side on a printer build plate — **Fig. 8**Revisions 1 through 4, left to right.

05 / VERIFICATION

A load cell settles the argument

I wrote a three-stage verification plan: fingertip force at a load cell, then friction coefficient, then static payload. For the first stage, each finger hung under an inverted Loadstar load cell. A NEMA 17 stepper on an A4988 driver wound the line onto a 31.25 mm capstan, curling the finger up into the cell. A Python script logged force to CSV and a MATLAB script turned each run into a force-time plot.

CAD model of the test rig: a plate on threaded rods holding a load cell above a hanging finger — **Fig. 9**Test rig in CAD.

Photo of the built test rig on a steel table with a stepper motor, Arduino, and load cell, with an industrial robot behind it — **Fig. 10**The built rig on the shop floor.

Fig. 11Peak fingertip force for each design and routing, one run each. Rev 2 with looped routing was strongest but can only push inward. Rev 1 looped was the best two-way design.

Rev 2 hit the highest peak at 2.4 N, but it slid off the load cell and can't apply force outward. Rev 1 with looped routing reached 1.4 N and works in both directions, so it became the recommended baseline. The handoff report also lists what the test can't tell you yet: single runs, friction in printed joints, and line stretch. It recommends at least three repeats per configuration before anyone relies on the numbers.

06 / OUTCOME

Handed off early and under budget

$11.2k

secured from the pitch

3 wks

early

$5k

under budget

force-test configurations

I briefed HIROTEC executives on the design, which compared well against a multi-month effort by the parent company's team in Japan, then wrote a formal handoff package: design rationale, test procedure and code, raw data, and next steps for the friction and payload phases.

Along the way I also flagged process gaps in weld verification and robot programming. The fixes I proposed are projected to cut verification time by up to 50% and to stop errors that kept coming back because of missing documentation.

Also at HIROTEC

Shop-floor fixes

Smaller parts I designed in NX and printed for the production floor, most of which went into daily use.

CAD model of a U-shaped tray with angled walls — Reamer trayCatches wire scrap in the MIG cell so it can't jam the robot's belt. In use.

CAD model of a two-piece cylindrical pipe coupler — Vacuum couplerReplaced melted tape on a weld robot's fume extractor line.

CAD model of an L-shaped sensor bracket — Weld wire sensor holderSenses low wire in the MIG cell, part of a push toward running it with no operator.

CAD model of a yellow gripper pad covered in small round studs — TPU gripper padsStudded flexible pads for a cobot. Running on a production line.

Color-coded CAD model of a cylindrical robot riser with a top plate — Mini robot riserPrintable scale model of a riser for showing customers.

CAD model of a long angled bracket — iPad mountHolds tablets for a project to streamline riser checks.

CAD model of a round keychain with a raised letter h — Company keychainTwo-color logo keychain for the company picnic.

Slicer view of a full printer build plate covered in green round keychains — Batch printingA full build plate of them, set up and printed on the shop's Bambu printer.